CodecFlow CN

机器人开发最劝退人的环节常常是前期配置，而SimArena直接在浏览器里运行，开发者打开网页就能开始模拟机器人行为。 simarena.ai

机器人开发正逐渐变成一件可以独自起步的事：过去，通常需要一个经费充足的团队来搭建场景、采集数据、训练 policy，并观察机器人的执行效果；现在，一名工程师、一个浏览器和一条低价机械臂，就能跑通这一套流程。

The visions of @codecopenflow and @peaq align, as we’re focusing on different parts of the same stack. At the end of the day, it goes down to democratizing access. Access to capital, access to resources, access to systems, access to services, knowledge and tools. For machines and for humans working on and with these machines. This is a team that is serious about their mission and that’s building a product suite that is relevant. And it’s an honor for me to support it and to be a part of that journey. Looking forward to collaborating even closer with @unmoyai, @0xdetweiler, and the team. More of the above is coming to those who build. Stay tuned.

@codecopenflow @MartinElKhouri @peaq welcome👏 @MartinElKhouri

Once robots can plug into a model and start acting in the world, a second question shows up fast: who owns the machine, and how does it earn. @MartinElKhouri, CBO of @peaq, has been building exactly that layer. He's joining CodecFlow as an advisor on business and strategy.

behavior cloning很直观：向机器人演示一项任务，它会照着做一遍，但它学到的可能只是动作，而不是任务本身。正如图中的机械臂没有真正读取到“把方块放进碗里”这个目标，只是在复现一段曾经有效的动作，当桌上的碗被移开后，原本成功的动作轨迹便会失效。 因此，开发者需要通过模拟覆盖更多位置、角度和物体状态，让机器人在部署前学会应对这类变化。

Winner #6: @codecopenflow CodecFlow is an open-source platform for building robots and autonomous operators. It brings simulation, compute, and deployment together so teams can build, train, and ship robots from one place instead of stitching together a different tool at every step. Its browser-based simulator, SimArena, lets anyone build and test a robot with no hardware or setup. Read more:

代码上线前要测试，机器人更新部署到硬件前也应该先跑一轮模拟，看看行为是否符合预期。

@solanaaihq codec coded🤖@codecopenflow

Which Solana x402 coins do you support?👇 $PAYAI $AMIKO $DREAMS $ZAUTH $CODEC $CORAL $AVO $BREW $OPUS $DEXTER $SWTCH $ACE $FDRY $ZARA $HEY $ELIZA $JOBS $REL $WURK $DIRA $MOSS $SYRA $X4PAY $JTVO

如果开发者只在一个固定场景里训练机器人，模型可能记住的是场景细节，而不是任务本身。同一盏灯、同一张桌子、同一种摩擦、同一个物体颜色，都会变成训练分布的一部分，真实部署时，只要这些条件变了，动作输出就可能偏掉。所以训练时需要随机调整光照、摩擦、颜色和初始位置，让机器人在大量略有差异的场景里学习。 真实世界不会和模拟完全一致，但模型见过更多变化后，部署时就不容易被单一环境条件带偏。

booked an R1 EDU from the @UnitreeRobotics distributor here in Dubai the plan: improve the sim2real pipelines in SimArena, take browser built scenes, sensors, policies, datasets, and deploy them on real hardware most of the work is going to be on system ID UX. matching the URDF, actuator dynamics, and latency to the real R1 in the simulators, and making the flow as easy as possible going to document the whole thing on video. setup, sysid, the first transfer attempts, what breaks, what holds up. should be useful for anyone trying to close the sim2real gap

Open a browser, drag in a couple of robots, start collecting data.

先从搭建一个机器人场景开始：simarena.ai

有计划，但现阶段重点还是先把 SimArena做成机器人开发者愿意反复使用的基础工具，让其能更快搭场景、看数据、推进模拟和训练流程。商业化会围绕真实使用需求逐步展开，比如更大规模的仿真训练、团队协作、数据管理和工程集成等， solana:69LjZUUzxj3Cb3Fxeo1X4QpYEQTboApkhXTysPpbpump 也已经在 Solana 上线，和生态使用场景一起推进。

@CodecFlowCN 你们怎么产生收入呢 有计划吗

传统Isaac Sim场景搭建流程里，配置NVIDIA账户、RTX GPU、OpenUSD、URDF和传感器会消耗掉几天时间，SimArena则将机器人开发的启动成本降下来，让开发者先在网页里创建场景运行机器人，观察测试表现，等项目经过快速验证后再决定接入Isaac Sim做更复杂的模拟。

@0x0funky @codecopenflow 感谢方神一如既往看好codecflow！

滿認同AK所說的，下一個爆發的賽道，大概就是 Physical AI。 過去大家把資金集中在模型、算力、AI infra，但當 AI 開始走進真實世界，需求會開始往下游擴散：資料收集、場景驗證、供應鏈、機器人部署、真實環境 feedback loop。 所以我還是看好 @codecopenflow 會成為這波 Physical AI 裡很重要的資料收集項目，尤其是 SimArena 這個方向。 模型會越來越強，但要讓 AI 真正理解並操作現實世界，關鍵會變成誰能持續收集高品質的 real-world data，並把模擬、評測、任務互動和資料回流串起來。 Always DYOR。

every service in SimArena (chat, training, sim, generation, etc.) was running separate, each doing its own auth and billing today they all run through one layer. which means we can route discounts on inference and compute to holders and partners

几乎每一位机器人开发者都想过放弃的五个瞬间，大家中了几个？ simarena.ai